The Qube! worker can run either as a daemon (a "system service" on Windows), or a user process. When it runs as a daemon, it's said to be running in service mode. When it's run as a user process, it's said to be running in Desktop User mode.
Service mode vs. Desktop User mode
- the worker process is usually started at system boot time and runs as long as the system is up
- the worker process runs as either a Windows service or a daemon owned by the root user on OS X and linux.
- the worker process will run jobs under a user other than root or the system service. This user is determined by the proxy_execution_mode value:
proxy_execution_mode = proxymeans it will always authenticate as the user defined in proxy_account.
proxy_execution_mode = usermeans it will always authenticate as the user who submitted the job.
Desktop User mode:
- the worker process is started by a logged in user, usually when that user logs in, and is killed when the user logs out
- the worker process is owned by the logged in user
- the worker process does not re-authenticate, and all jobs are run by the user who owns the worker process
- the worker process has full access to the screenspace
Worker authentication when running in Service mode
When the Worker launches a remote job as dispatched by the Supervisor, it can potentially create several processes, all controlled by the Worker. Since Qube is designed to emulate a user executing jobs on a remote host, the Worker will have to run these processes as some user.
Unix & OS X
Under Unix-based operating systems, the Worker does a setuid in order to switch user identities before starting the process. One implication of this is that the user's shell environment is not set up (tcshrc or equivalent is not run), so it will not be identical to an actual login by that user.
On Windows, this is handled quite differently. Under Windows, any process attempting to impersonate another user will have to provide certain security information, including the user's encrypted password.
To enable the Worker to provide this information every time it creates a new process, the information is stored in the worker's configuration, either in the worker's local qb.conf or the central worker configuration file stored on the supervisor.